FR2988618A1 - ALTITUDE ESTIMER FOR MULTI-ROTOR ROTOR SAIL DRONE - Google Patents

Abstract

The drone includes altitude determination means (134) with an estimator (152) combining measurements of an ultrasonic telemetry sensor (154) and a barometric sensor (156) to provide an absolute altitude value of the drone in a landmark. The estimator includes a predictive filter (152) incorporating a representation of a dynamic model of the drone for predicting the altitude from the engine controls (158) and periodically recalibrating the prediction based on the signals delivered by the telemetry sensors ( 154) and barometric (156). Validation means analyze the echoes returned and possibly modify the parameters of the estimator and / or allow or invalidate the signals of the telemetric sensor. The echo analysis also allows to deduce the presence and the configuration of an obstacle in the range of the telemetric sensor, to apply if necessary an appropriate corrective action.

Description

The invention relates to rotary wing drones such as quadcopters and the like. These drones are provided with multiple rotors driven by respective engines controllable in a differentiated manner to control the drone attitude and speed. A typical example of such a drone is the AR.Drone of Parrot SA, Paris, France, which is a quadricopter equipped with a series of sensors (accelerometers, three-axis gyrometers, altimeter), a front camera capturing an image of the scene to which the drone is directed, and a vertical aiming camera capturing an image of the terrain overflown. WO 2010/061099 A2 and EP 2 364 757 A1 (Parrot SA) describe such a drone and its driving principle via a phone or multimedia player with touch screen and built-in accelerometer, for example a cell phone of type iPhone or a player or a multimedia tablet type iPod Touch or iPad (trademarks of Apple Inc., USA). The drone is piloted by the user by means of signals emitted by the inclination detector of the aircraft, inclinations which will be replicated by the drone: for example, to advance the drone the user inclines his aircraft according to his axis. pitch, and to deport the drone right or left it inclines the same device relative to its axis of roll. In this way, if the drone is commanded to tilt or "dive" downward (tilt at a pitch angle), it will progress forward with a speed that is higher as the tilt important; conversely, if it is controlled so as to "pitch up" in the opposite direction, its speed will gradually slow down and then reverse by going backwards. In the same way, for an inclination control along a roll axis the drone will lean to the right or left, causing a linear displacement in horizontal translation to the right or to the left.

The user has other commands displayed on the touch screen, including "up / down" (throttle control) and "right swivel / left swivel" (swiveling of the drone around its yaw axis). ). The drone is also equipped with a fixed-point control: when the user releases all commands from his remote control device, the drone comes to a stop at the fixed point and stabilizes itself there fully automatically. The invention relates more particularly to the evaluation of the altitude at which the drone moves, and its vertical stabilization during its evolution, from the estimated altitude. "Altitude" means the value of the instantaneous position of the drone in the vertical direction, considered in a fixed landmark such as a Galilean landmark, the zero altitude of which corresponds to the position of the drone on the ground, at the time of its lift-off. This "altitude" is therefore an absolute size. The drone such as the AR.Drone described in the aforementioned documents is equipped with an ultrasonic rangefinder (US) comprising an electroacoustic transducer for transmitting and receiving ultrasound. This transducer emits a brief burst of ultrasound a few tens or hundreds of microseconds, then waits for the return of the acoustic echo sent after reflection on the ground. The time interval between the emission of the salvo and the reception of the echo makes it possible, knowing the speed of the sound, to estimate the length of the acoustic path traveled and thus to evaluate the distance separating the drone from the reflecting surface. In reality, since the beam of the US sensor is relatively wide (typically a cone of approximately 55 ° aperture), the transducer most often receives a multiplicity of echoes, and discriminates among these echoes that which corresponds to the point the closest. This measurement is iterated at short intervals, with a frequency of recurrence typical of ultrasound bursts of 25 Hz. The result provided by this US rangefinder, hereinafter "distance", is therefore a relative quantity, depending on the relief of the terrain overflown by the drone. The distance measured may differ from the altitude (in the sense indicated above), especially when the drone comes to pass over an obstacle, for example if it overflies, at constant altitude, a table or a wall During this flight, the distance measured by the US rangefinder will decrease sharply, even though the altitude will not have changed. If we only stick to the indications of the range finder, we are therefore likely to have the drone perform "field monitoring", which is not the goal, especially in rough terrain when we want only maintain the altitude at a stable value. The invention aims to solve a number of problems resulting from this phenomenon, as well as other disadvantages peculiar to US telemetric sensors. These sensors have the following characteristics: - the measurement produced is only a relative measure of altitude (distance measurement, telemetry); - in real situation, the measurement is very parasitized, because of multiple echoes returned from the ground, more or less reflective terrain and frequent disappearance of the signal for example when the drone flies over an absorbing ground (scrub ... ); the range is limited, about 6 m in the case of the AR.Drone described in the aforementioned documents, and beyond this value the telemetric signal disappears suddenly; - On the other hand, the measurement is very fast, it can be repeated at a high frequency (typically 25 Hz), and its accuracy is excellent, of the order of a few centimeters on a measurement scale ranging from a few tens of centimeters to several meters. .

To overcome these disadvantages, it is possible to use in combination with the rangefinder sensor another type of sensor, namely a pressure sensor, or barometric sensor. A barometric sensor makes it possible to measure pressure variations during flight, variations that are correlated with altitude variations. It is thus possible to obtain an absolute measurement of the altitude by integrating these variations from a zero altitude at the time of takeoff. A barometric sensor has the following characteristics: - it provides an absolute measurement, independent of the terrain overflown; - it is usable at high altitude, without upper limit; - On the other hand, it is slow and imprecise, insofar as it is necessary to integrate pressure variations; - Furthermore, it is subject to aerodynamic disturbances, especially at low altitude due to the ground effect, when the drone rotors produce significant turbulence making unusable the signals delivered by the pressure sensor.

In fact, a simple combination of the signals delivered concurrently by two telemetric and barometric sensors does not always make it possible to obtain a correct estimate, evaluated continuously, of the altitude of the drone.

However, if one wishes in particular to enslave the drone to stabilize it in the direction of the vertical, in fixed point or piloted flight, it is necessary to have a continuous and reliable way of a measurement of the altitude, of a way that does not risk disrupting the servo loops controlling the stabilization of the drone.

Thus, the object of the invention is to propose a drone comprising altitude determination means provided with validation algorithms and possibly readjustment of the estimated altitude measurement, which makes it possible to solve this difficulty in all circumstances, in particular when the drone moves over a terrain whose nature and singularities are a priori unknown and unpredictable. The invention proposes for this purpose a rotary-rotor rotary-wing drone driven by respective engines controlled by application of differentiated engine commands for steering the drone in attitude and speed, and comprising altitude determination means capable of delimiting set an absolute altitude value of the drone expressed in an absolute landmark. Characteristically, the altitude determination means according to the invention comprise: a telemetric sensor capable of estimating a relative distance separating the drone from a surface returning echoes of ultrasonic pulses emitted by the drone; a barometric sensor capable of delivering a signal of altitude variation of the drone; and an estimator receiving as input the signals delivered by the telemetric sensor and the barometric sensor and combining these signals to output said absolute altitude value of the drone. In addition, means for validating the signal delivered by the telemetric sensor analyze the returned echoes and, depending on the result of this analysis, modify accordingly the parameters of the estimator and / or allow or inhibit the application to the estimator of the signal delivered by the telemetric sensor. Finally, obstacle detection means analyze the echoes returned and, depending on the result of this analysis, deduce the presence and configuration of an obstacle in the range of the telemetric sensor and apply to the estimator an action corrector adapted to compensate the effect of the obstacle on the relative distance estimated by the telemetry sensor. Very advantageously, the estimator comprises a predictive filter incorporating a representation of a dynamic model of the drone, this filter being able to make a prediction of said absolute altitude value of the drone from said engine commands and to periodically readjust this prediction according to the signals delivered by the telemetric sensor and the barometric sensor.

This predictive filter is in particular a four-state filter, comprising: said absolute altitude value, counted with respect to a take-off position of the drone; a vertical speed component of the drone; a bias of said engine commands with respect to said dynamic model of the drone; and a bias of the barometric sensor.

According to various advantageous subsidiary characteristics: the validation means comprise means for estimating the quality of said echo-return surface, as a function of the number of echoes received concurrently for the same emitted pulse, and / or as a function of the dispersion between the distance values obtained for a temporal succession of pulses emitted by the telemetric sensor; the validation means comprise means of rejection of the distance values estimated by the telemetric sensor, as a function of a coherence analysis of a plurality of distance values obtained for a temporal succession of pulses emitted by the cape; telemetric transmitter, and / or as a function of the difference of a current distance value with respect to a distance prediction obtained from a plurality of distance values obtained for a temporal succession of pulses emitted by the sensor telemetry; - The rejection means operate with respect to variable thresholds, a function of a quality estimate of the surface returning the echoes; means are further provided for calculating an altitude offset value at the time the rejection means reject the estimated distance values, and for adding this offset value to the estimated distance value when the rejection no longer reject the estimated distance values; the obstacle detection means comprise means for detecting, in the distance values obtained for a temporal succession of pulses emitted by the telemetric sensor, two consecutive jumps of opposite sign with similar distance values obtained before and after these two jumps, advantageously with means for calculating an altitude shift value at the moment of the first jump, and for adding this offset value to the estimated distance value at the time of the second jump; the obstacle detection means comprise means for detecting an oscillation, between two values, of the distance values obtained for a temporal succession of pulses emitted by the telemetric sensor when the drone is in a fixed point state; the obstacle detection means comprise means for smoothing the distance values successively obtained in the presence of a detected oscillation; the obstacle detection means comprise means able to detect a loss of the signal delivered by the telemetric sensor due to the absence of returned echo, to calculate an altitude offset value at the time of this loss. signal, and to apply this offset value as a telemetric signal at the input of the estimator. In particular, in case of loss of telemetry signal: i) if the altitude of the drone is lower than a given threshold altitude, the recorded offset value is applied to the input of the estimator if this recorded value is present, a predetermined value being applied in the opposite case, and ii) if the altitude of the drone is greater than said given threshold altitude, the recorded offset value is applied unchanged at the input of the estimator. An embodiment of the device of the invention will now be described with reference to the appended drawings in which the same numerical references designate identical or functionally similar elements from one figure to another. Figure 1 is an overview showing the drone and associated remote control device for remote control.

Figure 2 is a block diagram of the various control organs, servo and assisted steering of the drone. Figure 3 illustrates different configurations of terrain overflight by the drone at constant altitude. Figure 4 is a flowchart of the different functions of validation of the US sensor measurements and application of possible corrective measures. Figure 5 is a timing chart showing the successive states of the Boolean indicator representative of the quality of the soil as a function of the number of echoes received by the US sensor.

Figure 6 shows two timing diagrams illustrating the variations of the raw measurement delivered by the US sensor, and the number of simultaneously received echoes, during the overflight of a terrain having variable characteristics of absorption of the transmitted US signal. FIG. 7 is a timing diagram showing the successive states of the Boolean indicator representative of the consistency of the measurements delivered by the US sensor as a function of the values of these measurements successively acquired and predicted. Figure 8 illustrates the update of the offset applied to the raw measurement delivered by the US sensor to the passage of an obstacle, to prevent the drone does ground tracking during the overflight of this obstacle. Figure 9 illustrates a phenomenon of altitude loss by the drone following the passage and the removal of an obstacle, because of the inclination of the drone during this flight. Figure 10 is a timing chart showing the variation of the different signals in the obstacle clearance configuration shown in Figure 9, as well as the successive states of the Boolean signal rejection flag delivered by the US sensor. Figure 11 is a timing chart showing how to handle the offset applied to the raw measurement output from the US sensor in the obstacle clearance configuration of Figures 9 and 10.

FIG. 12 illustrates a phenomenon of instability of the measurement delivered by the US sensor of the drone when it is flying over a fixed point on the border of an obstacle. FIG. 13 is a timing diagram showing the variations of the raw measurement delivered by the US sensor in the fixed point configuration at the obstacle boundary of FIG. 12, as well as the successive states of the Boolean signal rejection indicator delivered by FIG. the US sensor. FIG. 14 is a timing diagram illustrating the application of the US sensor measurement correction algorithm in the configuration of FIGS. 12 and 13. FIG. 15 illustrates the signal loss phenomenon delivered by the UAV sensor in US case of sudden increase in distance from the ground. FIG. 16 is a timing diagram illustrating the application of the algorithm for correcting the measurements and for resetting the values estimated by the configuration of FIG. 15. In FIG. 1, the reference 10 generally designates a drone, for example a quadricopter such as the aforementioned AR.Drone described in WO 2010/061099 A2 and EP 2,364,757 A1, as well as FR 2,915,569 A1 (which describes in particular the system of gyrometers and accelerometers used by the drone) and EP 2 431 084 A1 (which describes in particular the manner of controlling predetermined trajectories). The drone 10 comprises four coplanar rotors 12 whose engines are controlled independently by an integrated navigation system and attitude control. Inertial sensors (accelerometers and gyrometers) make it possible to measure with precision the angular velocities and attitude angles of the drone, ie the Euler angles describing the inclination of the drone with respect to a horizontal plane of a fixed terrestrial landmark. The drone is equipped with a first frontal-aiming camera, which provides an image of the scene to which it is oriented, as well as a second camera with a vertical aim, pointing downwards to capture successive images of the camera. terrain overflown and used in particular to evaluate the speed of the drone relative to the ground in combination with the accelerometer data, thanks to software that estimates the movement of the scene captured by the camera from one image to the next and applies to this estimated displacement a scale factor as a function of the measured altitude (this technique is described in detail in EP 2 400 460 A1 (Parrot SA), to which reference may be made for further details). An ultrasonic range finder underneath the drone and an onboard barometric sensor also provide measurements which, when combined, give an estimate of the altitude of the drone relative to the ground. More specifically, the ultrasonic rangefinder comprises an electro-acoustic transducer 14 generating towards the ground 18 a brief ultrasound burst (of a few tens or hundreds of microseconds) into an approximately conical beam 16, with an opening angle of 20 degrees. about 55 °. The reflection of this beam on the ground 18 produces acoustic echoes which are received by the transducer 14 and analyzed so as to estimate the length of the acoustic path traveled, as a function of the time interval separating the emission of the burst from the reception of the first echo and speed of sound, and thus evaluate the distance separating the drone 10 from the ground 18. This distance measurement is, in a manner characteristic of the invention, intersected with a measurement of the pressure variations produced by the variations of altitude of the drone during its evolution, in the way that will be exposed below.

The drone 10 is controlled by a remote remote control device 22 provided with a touch screen 24 displaying the image embedded by one of the cameras of the drone, with in superposition a certain number of symbols allowing the activation of control commands by a simple touch of the finger 26 of a user on the touch screen 18. The device 22 is provided with inclination sensors to control the attitude of the drone by printing the device corresponding inclinations. For the bidirectional exchange of data with the drone, it is also provided with radio link means, for example Wi-Fi network type (IEEE 802.11) or Bluetooth (registered trademarks). As indicated in the introduction, the remote control device 22 is advantageously constituted by a multimedia touch screen phone or scanner with an integrated accelerometer, for example an iPhone-type cell phone, an iPod Touch-type music player or a multimedia tablet. of type iPad, which are devices incorporating the various control organs necessary for the display and detection of control commands, the viewing of the image captured by the front camera, and the bidirectional exchange of data with the drone via Wi-Fi or Bluetooth. The piloting of the drone 10 consists in making it evolve by controlling the engines in a differentiated manner to generate, as the case may be, movements of: a) rotation about a pitch axis, to advance or retreat the drone ; and / or b) rotation around a roll axis, to shift the drone to the right or to the left; and / or c) rotation about a yaw axis, to rotate the main axis of the drone to the right or to the left; and / or (d) downward or upward translation by changing the rate of the gases so as to reduce or increase the altitude of the rotor, respectively.

When these control commands are applied by the user from the apparatus 22 (controlled pilot configuration, controlled by a user), the commands a) and b) are obtained by inclinations of the apparatus: by for example, to advance the drone simply tilt the device by leaning forward, to deport to the right it is sufficient to incline the device by leaning to the right, etc. The commands c) and d), as for them, result from actions applied by contact of the finger of the user on corresponding specific areas of the touch screen (reference may be made to WO 2010/061099 A2 and EP 2,364 757 Al supra for more details on these aspects of the system).

The drone also has an automatic and autonomous hover stabilization system ("fixed point" configuration, autopiloted), activated especially when the user removes his finger from the touch screen of the device, or automatically at the end of the take-off phase, or in case of interruption of the radio link between the aircraft and the drone. The 2 9886 18 11 drone then goes to a state of levitation where it will be immobilized and stabilized automatically, without any intervention of the user. Control and control of the engines of the drone 5 We will now expose the way in which are developed the steering instructions of the engines of the drone, with reference to Figure 2 which is a functional block diagram of the various controls and servo control of the drone . It will be appreciated that although these diagrams are presented as interconnected circuits, the implementation of the various functions is essentially software-based, this representation being illustrative only. In general, as illustrated in FIG. 2, the control system involves several nested loops for the control of the horizontal velocity, the angular velocity and the attitude of the drone, in addition to the variations of altitude. The most central loop, which is the angular velocity control loop 100, uses, on the one hand, the signals supplied by gyrometers 102 and, on the other hand, a reference consisting of angular velocity instructions 104. different information is applied at the input of a stage 106 for correcting the angular velocity, which itself controls a motor control stage 108 in order to separately control the speed of the various motors in order to correct the angular velocity of the drone by the combined action of the rotors driven by these engines. The angular velocity control loop 100 is embedded in an attitude control loop 112, which operates from the indications provided by the gyros 102, by accelerometers 114 and by a magnetometer 116 giving the absolute orientation of the drone. in a terrestrial geomagnetic landmark. The data from these different sensors 30 are applied to a PI type attitude estimation stage 118 (Proportional-Integrator). The stage 118 produces an estimate of the real attitude of the drone, applied to an attitude correction stage 120 which compares the real attitude with angle commands generated by a circuit 122 from commands directly applied by the driver. 124 (configuration of "controlled steering") and / or data generated internally by the autopilot of the drone via the horizontal speed correction circuit 126 for holding the drone stationary ("fixed point configuration "autopiloted"). The possibly corrected instructions applied to the circuit 120 and compared to the actual attitude of the drone are transmitted by the circuit 120 to the circuit 104 to control the motors appropriately. In summary, from the error between the setpoint (applied by the user and / or generated internally) and the measurement of the angle (given by the attitude estimation circuit 118), the loop 112 of Attitude control calculates a setpoint in angular velocity using the corrector PI of the circuit 120. The control loop 100 of the angular velocity then calculates the difference between the previous angular velocity reference and the angular velocity actually measured by the gyrometers 102. The loop calculates from this information the various setpoints of rotational speed (and therefore of the upward force), which are sent to the motors 110 to perform the maneuver requested by the user and / or planned by the autopilot. of the drone. The horizontal speed control loop 130 uses a vertical video camera 132 and an altitude estimator circuit 134 (described in more detail below). A circuit 136 processes the images produced by the vertical camera 132, in combination with the signals of the accelerometer 114 and the attitude estimation circuit 118, to estimate by means of a circuit 138 the two components of the horizontal speed of the drone. The estimated horizontal speeds are corrected by the vertical speed estimate given by a circuit 140 and by an altitude value obtained from the estimator 134 and a correction / compensation circuit 142 of the altitude values in certain particular flight configurations of the drone (this aspect will also be described in more detail below). With regard to the vertical displacements of the drone, the user 124 applies commands to a circuit 144 for calculating altitude readings, which are applied to a circuit 146 for calculating the ascending speed reference V, via the altitude correction circuit 148 receiving the altitude estimation given by the circuit 142. The calculated climbing speed V is applied to a circuit 150 which compares this target speed with the corresponding speed estimated by the circuit 140 and consequently modifies the control data of the motors (circuit 108) so as to increase or decrease the speed of rotation simultaneously on all the motors so as to minimize the difference between the ascending velocity of reference and measured rate of climb.

The invention relates more precisely to the manner in which the altitude estimation circuit 134 operates and the correction / compensation circuit 142 for estimating the altitude value. First of all, the manner in which the altitude estimation circuit 134. operates. This circuit implements a "Kalman filter" type state estimator, which is an infinite impulse response filter which estimates the states. of a dynamic system (the drone in this case) from a series of input measurements. The general principles of this technique can be found for example in R. E. Kalman, A New Approach to Linear Filtering and Prediction Problems, Transactions of the ASME - Journal of Basic Engineering, Vol. 82 (1960).

In the present case, the Kalman filter receives as input: the signal delivered by the US sensor 154, the signal delivered by the barometric sensor 156, and at 158 the setpoint (hereinafter referred to as "PWM setpoint") of rate of rise produced by the vertical speed correction circuit 150 Vz and applied to the drone motor (circuit 108): the PWM setpoint represents the vertical thrust force resulting from the simultaneous control of the four engines of the drone. The dynamic Kalman filter system consists of a set of equations that describe the behavior of the drone. Such modeling is described in particular in EP 2 431 084 A1, to which reference may be made for further details. The Kalman filter 152 operates in two phases, with successively: a prediction phase, performed at each iteration of the filter: this phase consists in predicting the altitude of the drone at the current instant using a piece of the PWM setpoint and secondly dynamic modeling of the drone; a correction phase, which consists in correcting the prediction using the current measurements delivered by the US 154 and barometric sensors 156. This step is not necessarily carried out at each iteration because, as will be seen below , US sensor measurements may not be available all the time. The Kalman filter uses and estimates four states, namely: - the altitude of the drone relative to its starting point (position at the time of take-off), - the vertical speed Vz, - the bias of the PWM setpoint, and - the bias of the pressure measurement delivered by the barometric sensor 156 The state corresponding to the bias of the PWM setpoint represents the error of the thrust model of the drone: the thrust pattern can indeed vary according to various parameters such as the weight of the drone, the aerodynamic efficiency of the propellers (this efficiency depends on the condition of the propellers and the air pressure), etc. The bias state of the PWM setpoint serves to recalibrate the model in flight using the measurements provided by the two sensors, which in particular makes it possible to be more precise when these measurements disappear. The bias state of the pressure measurement makes it possible to correct the drifts of the barometric sensor. These appear mainly at take-off, when the engines are started and during the first seconds of flight with the rise in temperature. These drifts may also appear when the pressure environment is changed in other circumstances, for example when the drone moves inside a room and a person opens a window of this room, etc. This state of bias of the pressure measurement is estimated from the measurements of the US 15 sensor. When the US measurement is no longer available, for example when the drone is gaining altitude beyond the range of the US sensor, or no longer receives an echo for another reason (soil too absorbing) the state of bias of the pressure measurement is no longer estimated, it is kept at a constant value.

To recalibrate the states of the filter, it uses three observations: - the distance measurement, given by the sensor US 154, possibly corrected for a value called offset when passing an obstacle (this aspect will be developed in detail lower). It should be noted that this observation is not always available: under optimal conditions, on smooth ground and at low altitude, the measurement is provided at a recurrence frequency of 25 Hz, ie 1 time out of 8 with respect to the frequency recalculating the control data of the drone, which is typically 200 Hz. On the other hand, the measurement disappears as soon as the altitude exceeds the radius of action of the ultrasonic range finder, typically beyond 6 m, and it can moreover be rejected regularly during the passage of obstacles or on a chaotic ground (this description will be described in more detail below, in particular the way in which decisions are made about rejection or not of the measurement delivered by the US sensor) ; the measurement of pressure, given by the barometric sensor 156, corrected by its own offset. This measurement is generally present at each iteration, but it is deactivated when the drone is very close to the ground, typically less than 20 cm from the ground, taking into account the disturbing blast effects that strongly parasitize the pressure measurement; and zero altitude measurement, which is used only when the drone is not in flight; it is a virtual measurement that ensures that the altitude value delivered to the navigation circuits remains set to zero.

Validation of the data delivered by the ultrasonic range finder This aspect of the signal management delivered by the US rangefinder is intended to i) decide whether, at a given moment, the US measurement is present and if it is exploitable, and ii) if necessary to update the offset value when passing an obstacle. The characteristics of the US measurement being strongly linked to the more or less reflective nature of the ground, and being moreover very different depending on whether the drone is in a fixed point or in horizontal displacement, several algorithms are used to manage the different scenarios. - to be met.

Figure 3 illustrates these different configurations of flying over the terrain by the drone. Initially, the drone is placed on the ground and motionless (position 160); the altitude value is then forced to zero.

The drone then rises above the ground (position 162), then performs a movement of horizontal movement, at constant altitude (positions 164 to 172). The US sensor does not give an indication of the absolute value of the altitude, it only gives a representation of the distance d separating the drone from the ground 18. In the case of a horizontal ground, regular and well reflective, this distance can be evaluated without ambiguity. On the other hand, when the drone flies over, for example, in a low reflective soil (brushwood, etc.), the echoes returned are either non-existent or erratic, leading to a strongly noisy measurement. Another aspect is the passage of an obstacle, for example the flying over an outside wall or an indoor table, as illustrated in 168 in Figure 3: if the obstacle is reflective, the rangefinder will measure a distance dl (position 168) much lower than those d2 and d3 measured before and after the overflight of the obstacle (positions 170 and 172), even as the drone moves at constant (absolute) altitude. The crossing of the obstacle, revealed by a sudden decrease of the US sensor signal, must therefore be offset by an offset or "offset" applied to the US measurement during the crossing of the obstacle.

Thus, it is necessary: - to evaluate the quality of the signal received to validate or invalidate the measurement delivered by the US sensor - and, if this measurement is validated, to manage the offset that must be added thereto.

These functions for validating the measurements and for applying any corrections are illustrated on the flowchart of FIG. 4. The analysis of the US signal is of course carried out only if the sensor delivers a signal (test 200), that is, ie: i) 1 cycle over 8 cycles of data recalculation, and ii) if at least one echo is detected in the time window corresponding to the range of the US sensor.

It is then necessary to estimate the quality of the ground overflown by the drone (blocks 202 and 204), more precisely the quality of reflection of the echoes emitted by the US transducer, which reflection is very good on a smooth ground, but very parasitic on a more diffuse soil.

To account for these variations, the measurement noise setting of the Kalman filter will be adapted according to a Boolean good / bad indicator of soil quality (test 206): for example the noise level of the filter will be adjusted to 5 cm (block 208) on a clean, good quality floor, while it will be adjusted to 15 cm (block 210) on a floor of poor reflective quality. As a result, on a clean ground the filter will accurately adjust the estimated altitude on the distance measurement provided by the US range finder, while it will make it less confident, and thus filter more effectively the noise on a ground poor quality. This Boolean good / bad soil quality will also be used to modify the rejection thresholds for subsequent validation or invalidation of the measurement, and will also be used for offset management - these aspects will be described below. The overall estimate of the soil quality involves the merging of two specific Boolean indicators, one resulting from the analysis of the number of echoes (block 202), the other of the more or less stable character of the measurements successively. produced by the US telemeter (block 204). The soil will be declared good (test 206) if the two Booleans correspond to a good soil; it will be declared bad otherwise, that is, if at least one of the two Booleans considers the soil to be bad.

More specifically, the first criterion for evaluating soil quality (block 202) is to evaluate the nature of the soil as a function of the number of echoes reflected and captured by the transducer. In practical terms, the distance measurement of the US telemetry sensor corresponds to the return time of the first echo. But many other echoes are also received and recorded by the transducer, and the number of echoes received may be an estimate of the quality of the soil: for example, on a net soil usually only one or two echoes are received, whereas on a chaotic ground (bushes or obstacles), the number of echoes can be up to six or seven.

This measurement of the number of echoes being relatively noisy, it is first filtered using a first-order recursive filter. The algorithm then applies hysteresis thresholding (with, for example, a high threshold with 3 echoes and a low threshold with 2 echoes) to obtain the first soil indicator boolean. Figure 5 illustrates the result thus obtained during an example of flight in a configuration where the drone alternately flies over smooth ground and bushes. The second criterion of evaluation of the quality of the soil (block 204) is based on the stability of the successive measurements of distance delivered by the US sensor Indeed, in certain cases, the number of echoes is not sufficient to rule on the quality of the soil. Thus, in FIG. 6 there is illustrated an example showing the simultaneous variations of the raw measurement delivered by the US sensor and the number of echoes received, during the overflight of a field having variable characteristics of signal absorption. US issued. It can be seen that the measurement given by the US sensor is very noisy, a sign of poor reflective soil, even though the number of echoes remains low and is not sufficient to declare the soil "bad" on the basis of the only first Booster (function of the number of echoes received). To take this into account, a second Boolean of soil quality is calculated by observing the dispersion of the measurement: at each new measurement delivered by the US sensor, the algorithm calculates the difference (in absolute value) with respect to to the previous measure. This value is then filtered by a low-pass filter (the same as for filtering the number of echoes), then thresholded. A value for example greater than 40 mm will lead to give the value "bad" to the second boolean. This value of 40 mm is chosen to correspond, with a recurrence of measurements of 25 Hz, to a vertical speed of 8 m / s, a speed never reached by the drone, which minimizes the incidence of the bias that could appear when the drone has a non-zero vertical speed. In practice, so that the passage of an obstacle does not trigger the test, only the dispersions are retained, for example less than 50 cm; a higher dispersion will be considered as the result of jumping an obstacle and not as a revealing dispersion of poor quality soil. After evaluating the quality of the soil, the algorithm decides whether the measurement delivered by the US sensor should be validated or rejected.

We will describe a validation test (steps 212 to 216 in Figure 4) to meet the following requirements: - use only the measurement produced by the US sensor and not the prediction of the filter, to gain robustness and not to reject the US measurement in cases where it is good and the model of the filter is in fault (for example in a configuration with a side wind); - to present on several iterations a memory effect: indeed, in cases where the detection is very disturbed (bushes, etc.), a simple detection of a jump in the value delivered by the sensor is not sufficient, because it may lead to accepting this value too quickly when the measurement is not stable, so it is better to check the consistency of the measurement over a larger number of points; - not being sensitive to the vertical speed of the drone: to quickly reject any disturbance, it is indeed necessary to have fairly rigorous thresholds on the measured quantity, which must be independent of the vertical speed to avoid any risk of rejection of the measurement in the phases of rise and descent of the drone. The coherence test of the measure provides (block 212 in Figure 4) to maintain a history of the last five measurements of the US sensor and perform on them a linear regression. The consistency of the five points between them is then calculated for example by summing the deviations of each point from the regression line calculated on the basis of the five previous measurements. If the five points recorded in the history are not coherent with each other (too much dispersion) or if the current measurement (measurement of the US sensor) is too far from the measurement predicted by the regression calculation, then the current measurement will be rejected. . Figure 7 illustrates an example showing on the same diagram the US measurement delivered by the sensor, the altitude predicted by the regression calculation and the validation / rejection result of the US measurement.

It can be seen that in the ascent phases as in the first seconds, the US measurements are not rejected: this comes from the fact that the regression takes into account the vertical speed (increase of the altitude). On the other hand, there is good rejection of the measure during a passage of an obstacle (around t = 99-100 s and 102-103 s), and acceptance again of the measure after it has been found a return to five coherent successive samples. This consistency test compares (block 214) i) the consistency of the five measurements recorded, and ii) the deviation from the current point with respect to the prediction. relative to two respective thresholds. Advantageously, these two thresholds are not constant and vary according to two parameters, namely the soil quality indicator (determined previously, as indicated) and the speed of movement of the drone. Thus: - on a net floor, the two thresholds are set at an average value giving a good rejection of the disturbances while ensuring a low risk of false obstacle detection: for example respective values of 50 mm (coherence of the five points) and 20 mm (deviation from the current point); - on a ground disturbed with the drone fixed point or with a low speed of translation (lower than 300 mm / s), the ultrasonic measurement is noisier, because of the nature of the ground: the adjustment of the filter of Kalman is adapted therefore, to avoid a too high rejection rate, and the thresholds above are set at 100 mm and 50 mm, respectively; - on a disturbed ground with the moving drone (travel speed greater than 300 mm / s), the US measurement is very likely to jump from one echo to another, and the confidence index is low. It is therefore necessary to reject it at the slightest doubt, and for this the threshold on the current point is chosen very low: the respective thresholds are then for example fixed at 100 mm and 10 mm. As a result of these various tests, the US measurement is finally rejected or validated (test 216 in FIG. 4). In case of rejection (block 218), the Kalman filter does not use the current US measurement, and predicts its prediction only from the motor control commands (input 158 in Figure 2) and the signal delivered by the sensor. barometric (156 in Figure 2).

In the opposite case (validated measure), if the US measurement returns to a valid state after being rejected, it is potentially offset by an offset if the previous rejection was the result of passing over an obstacle. This situation is managed in block 220 of Figure 4.

In any case, if the measurement is validated, the Kalman filter uses (block 222) the measurement delivered by the US sensor (input 154 in FIG. 2). An example of this case is shown in Figure 8. The passage of an obstacle corresponds to the anomaly detected at t = 91-94 s. At the moment of arrival on the obstacle (around t = 91 s), the offset is used to keep a constant estimate of altitude. After crossing the obstacle (at t = 94 s), the offset is recalculated from the first US measurement considered valid after rejection of the previous measurements. This makes it possible to prevent the drone from monitoring the terrain, especially if after the obstacle the ground is not at the same level as before the obstacle. Data correction / compensation in the presence of obstacles In various circumstances, distance measurement by the US sensor can be disturbed by obstacles. A first phenomenon is illustrated schematically in Figure 9, where we see that the drone in horizontal progression (successive positions 174 to 180) tends to lose altitude just after crossing the obstacle (180 position). This phenomenon is due to the fact that even after the drone has passed the obstacle (positions 180 and 182), the beam 16 remains oriented towards the latter because the drone is slightly inclined forward (to be able to print a horizontal speed to it ). As a result, the rangefinder sees the obstacle move away and converts a portion of the horizontal velocity into vertical velocity: the estimator then incorrectly estimates that the drone is climbing and applying a compensation offset which, in fact, will result in altitude loss of the drone (position 182). Figure 10 shows the variation of the different signals (measurement delivered by the US rangefinder, altitude delivered by the estimator and Boolean rejection of the US measurement) before, during and after the overflight of the obstacle in the configuration of Figure 9 : - when the drone approaches the obstacle (position 176 of Figure 9), the US measure shows a decrease, but this one is without consequence, because it corresponds to a period where the US measurement is rejected; - on the other hand, just after crossing the obstacle (position 180), the measurement is no longer rejected and the virtual altitude increase, of about 15 cm, due to the inclination of the drone estimator to apply a corrective measure by recalculating the offset to compensate for this (supposed) climb by modifying the setpoints so as to go down by 15 cm.

To overcome this phenomenon and to prevent the drone from constantly losing altitude, the invention proposes to apply an obstacle jumping and offset management algorithm, the effects of which are illustrated by the timing diagrams of the Figure 11 (which are homologous to those of Figure 10, but with application of the management algorithm according to the invention). This algorithm is activated when the drone detects the overflight and the end of the crossing of an obstacle, when two conditions are met: - presence on the US sensor measurement of two consecutive jumps of opposite sign (S + and S- in Figure 11) ), and - delivery by the US sensor of substantially equal distance values (at a predetermined threshold near) at the time of the first jump and the moment of the second jump (d1 and d2 in Figure 11). If these conditions are met, consider i) that an obstacle has been overflown and crossed, and ii) that the ground level after the obstacle is substantially the same as that before the obstacle. In such a case, the offset is no longer recalculated, but it is forced to take the value it had before crossing the obstacle. The effect of this correction can be seen in Figure 11 in X on the estimated altitude, where it can be seen that after crossing the obstacle, the altitude estimate resumes substantially the level it had before the approach and the crossing of the obstacle. In other words, at the moment of the obstacle exit, although the previously observed error on the measurement of 15 cm is present, this error is corrected afterwards, when the measurement of the US sensor is no longer rejected. The difference between the measurement of the US sensor and the altitude estimate is thus identical before and after the obstacle, and the phenomenon of altitude loss is eliminated. As a precaution, this detection and compensation algorithm is activated only when the soil quality is considered good, because of the large variations in the measurement of the US sensor noted at the passage of an obstacle of poor reflective quality. A second phenomenon produced by an obstacle occurs when the drone is in fixed point on the border of an obstacle. This situation is illustrated in 184 in Figure 12.

When the drone is in fixed point on the border of a net obstacle, for example a table edge, the measurement provided by the US sensor is very unstable and jumps permanently between two values, namely the distance of the drone to the table and the distance of the drone on the ground. Figure 13 shows these variations of the raw measurement, as well as the successive states of the Boolean indicator of rejection of the US signal. The coherence test explained above, which requires at least five successive measurements compliant to validate the measurement given by the US sensor, causes in such a case very frequent rejections, so that the drone does not maintain its altitude correctly fixed point : in fact, the evaluation of the altitude rests in this case essentially on the pressure sensor, but the accuracy of this one is not sufficient to ensure a suitable stabilization in fixed point. However, it is unfortunate to reject the US measurement in such a situation because, apart from jumps, the measurement is very stable and it is very easy to see the altitude variations of the drone, of the order of 50 cm in Figure 13. , corresponding to the performance of the pressure sensor. To overcome this second phenomenon, a specific algorithm is activated when the drone is in a fixed point (this flight condition being known and corresponding to an "autopiloted" state of the drone).

A jump is defined as a difference in the measurement of the US sensor, for example at least 50 cm. When such a jump is detected, the value of this jump is subtracted from the following measures recorded in the history of the last five bars. The history thus no longer contains the raw measurements, but the measures smoothed after the elimination of the jumps. If, in abstraction jumps, the measurement is stable, this produces a relatively uniform curve which does not trigger the rejection of the measurement by the consistency check algorithm. On the other hand, in the case of poor quality soil, the measurement of the US sensor is noisy and contains variations at all frequencies. In this case, even after the skips are removed, the result remains noisy and the consistency check algorithm continues to operate appropriately, to validate only the compliant measurements. The result obtained is shown in Figure 14: in the case of a jump, only the first measurement is rejected, and it is no longer necessary to wait for five further measurements to find a validation of the US measurement. In addition, the use of the obstacle jumper management algorithm ensures that the absolute reference of the ground is not lost: the offset of the US signal oscillates between two values but does not diverge, and the drone maintains its altitude correctly: in Figure 14, we observe a very low variation of altitude, not exceeding about 10 cm between the beginning and the end of the chronogram. A third phenomenon likely to disturb the operation of the US sensor is that resulting from a loss of any measurement delivered by this sensor, due to the sudden disappearance of any echo.

An example of this phenomenon is shown in Figure 15: the drone moves in a room at relatively low altitude, for example 1 m from the floor (positions 186 and 188), then through a window (position 190). If the room was in floor, it is then at a much higher altitude, for example 10 m from the ground. In this case, the altitude value delivered by the estimator continues to be integrated thanks to the pressure sensor, but the offset is no longer updated. However this offset is important because, by difference with the altitude of the drone, it makes it possible to calculate the distance with the ground, which is an important parameter in particular for the algorithms of vision used to analyze the image of the vertical camera used to evaluate the horizontal speed of the drone. When the US measurement is lost, the distance to the ground returned to the computer of the drone is the estimated altitude, corrected by the last calculated offset. In the case illustrated in Figure 15, after passing through the window (position 190) this distance to the ground will always be considered equal to 1 m, so that if the drone descends, for example, by 2 m (position 192), the distance to the ground after applying the offset will be -1 m, incoherent value. To mitigate this phenomenon, the invention proposes a specific algorithm for managing the offset executed when the drone is at high altitude (the "high altitude" being an altitude greater than the range of action of the US telemetry sensors, typically one altitude above 6 m). When the measurement of the US sensor has disappeared for a duration greater than for example 2 s, this means that the drone is far from the ground, and the value of the ground distance is evaluated from the stored offset: - if this distance is greater than 6 m, the operation is not changed and the value of the offset is recorded; - if it is less than 6 m, this means that it is not correct and that the drone has probably passed over a vacuum (crossing a window, passing over a hole, etc. .).

In the latter case, the value of the offset must be modified: - if there is a registered offset value that is appropriate, it is taken over and used; - In the opposite case, the distance to the ground is fixed at an arbitrary value, for example 6 m.

Figure 16 illustrates the results obtained during a drone flight in a building yard. Initially, the drone takes off from the ground. During the ascent phase (t = 15 s at t = 20 s), the ground distance remains within the range of the US telemetry sensor and the altitude is therefore properly evaluated by the estimator, with offset correction. obtained by US sensor signal analysis. From t = 20 s, the altitude of the drone approaches the limits of range of the US sensor and one observes losses of the US signal, which are repeated, until a total loss of the signal, to t> 32 s. After t = 33 s, as the altitude is greater than 4 m, the last offset value recorded at that time is retained and stored. Towards t = 37 s, the drone approaches a wall, which produces the return of some echoes, which then disappear when the drone moves away from this wall. After the reappearance / disappearance of the US signal due to the attachment of the wall, the previously stored offset is applied, which allows to keep a distance value to the ground substantially constant despite the sudden depression that the measurement of this value during of this hanging. Note that in the example of Figure 16 the distance to the ground could be estimated at the beginning of flight, which allows it to be restored after hooking the window. In the example of Figure 15 with takeoff directly from a window, there would have been no stored offset and the ground distance would have been arbitrarily set to 6 m.

Claims (15)

REVENDICATIONS1. A multi-rotor rotary wing drone (10) driven by respective engines (110) controlled by application of differentiated engine controls (158) to control the drone in attitude and speed, said drone having means for determining altitude (134), capable of delivering an absolute altitude value of the drone expressed in an absolute terrestrial reference, characterized in that the altitude determination means (134) comprise: a telemetric sensor (154), adapted to estimating a relative distance separating the drone from a surface (18) returning echoes of ultrasonic pulses emitted by the drone; a barometric sensor (156) capable of delivering a signal of altitude variation of the drone; an estimator (152), receiving as input the signals delivered by the telemetric sensor and the barometric sensor and combining these signals to output said absolute altitude value of the drone; and in that they furthermore comprise: means for validating the signal delivered by the telemetric sensor, able to analyze the returned echoes and, depending on the result of this analysis, modify accordingly the parameters of the estimator and or allow or inhibit the application to the estimator of the signal delivered by the telemetry sensor; and obstacle detection means able to analyze the returned echoes and, depending on the result of this analysis, to deduce the presence and configuration of an obstacle in the range of the telemetric sensor. and applying to the estimator a corrective action capable of compensating for the effect of the obstacle on the relative distance estimated by the telemetry sensor. 2. The drone of claim 1, wherein the estimator comprises a predictive filter (152) incorporating a representation of a dynamic model of the drone, this filter being able to make a prediction of said absolute altitude value of the drone to from said engine commands (158) and recalibrate this prediction periodically as a function of the signals delivered by the telemetric sensor (154) and the barometric sensor (156). The drone of claim 2, wherein the predictive filter is a four-state filter, comprising: said absolute altitude value, counted with respect to a take-off position of the drone; a vertical speed component of the drone; a bias of said engine commands with respect to said dynamic model of the drone; and a bias of the barometric sensor. 4. The drone of claim 1, wherein the validation means comprises means for estimating the quality of said echo-returning surface, as a function of the number of echoes received concurrently for the same emitted pulse. 5. The drone of claim 1, wherein the validation means comprise means for estimating the quality of said echo-returning surface, as a function of the dispersion between the distance values obtained for a temporal succession of pulses emitted by the telemetry sensor. 6. The drone of claim 1, wherein the validation means comprise means for rejecting the distance values estimated by the telemetric sensor, as a function of a coherence analysis of a plurality of distance values obtained for a succession. time of pulses emitted by the telemetry sensor. 7. The drone of claim 1, wherein the validation means comprise means for rejecting the distance values estimated by the telemetric sensor, as a function of the difference of a current distance value with respect to a distance prediction. obtained from a plurality of distance values obtained for a temporal succession of pulses emitted by the telemetric sensor. 8. The drone of claim 6 or 7, wherein the rejection means operate with respect to variable thresholds, a function of a quality estimate of the surface returning the echoes. The drone of claim 6 or 7, further comprising means for calculating an altitude offset value (OFFSET) at the time the rejection means rejects the estimated distance values, and for adding this offset value to the estimated distance value when the rejection means no longer reject the estimated distance values. The drone of claim 1, wherein the obstacle detection means comprise means for detecting, in the distance values obtained for a temporal succession of pulses emitted by the telemetric sensor, two consecutive jumps of opposite sign with similar distance values obtained before and after these two jumps. The drone of claim 10, further comprising means for calculating an altitude offset value (OFFSET) at the time of the first hop, and for adding said offset value to the estimated distance value at the time of the first jump. second jump. 12. The drone of claim 1, wherein the obstacle detection means comprise means for detecting an oscillation, between two values, distance values obtained for a temporal succession of pulses emitted by the telemetric sensor when the drone is in a fixed point state. 13. The drone of claim 12, wherein the obstacle detection means comprises means for smoothing the distance values successively obtained in the presence of a detected oscillation. 14. The drone of claim 1, wherein the obstacle detection means comprise means adapted to detect a loss of the signal delivered by the telemetric sensor due to the absence of returned echo, to calculate an offset value. of altitude (OFFSET) at the time of this loss of signal, and to apply this offset value as telemetric signal input of the estimator. 15. The drone of claim 14, in which, in the event of loss of telemetric signal: i) if the altitude of the drone is lower than a given threshold altitude, the recorded offset value is applied to the input of the aircraft. estimator if this registered value is present, a predetermined value being applied in the opposite case, and ii) if the altitude of the drone is greater than said given threshold altitude, the recorded offset value is applied unchanged at the input of the estimator.